ALL INPredictions

Sergey Brin, Google Co-Founder | All-In Live from Miami

Tue, 20 May 2025 02:37:00 +0000

Back to episodes
Sergey Brin Partly Right 00:17:31 techai

Over the next few years, foundational AI models will continue converging into fewer, more general-purpose models rather than proliferating into many specialized models for individual domains.

“things have been more converging... increasingly it's also just becoming one model... we do get a lot of oomph occasionally we do specialized models... but the trends have not gone that way.” View on YouTube
Explanation

The frontier AI landscape has not converged into fewer models: distinct major model families from OpenAI, Google, Anthropic, xAI, Meta, DeepSeek, and Alibaba continue to proliferate, but within any single lab's own lineup, general-purpose models have absorbed capabilities (coding, vision, agentic tool use) that once required separate specialized models, partially matching the prediction.

Sergey Brin Right 00:30:38 techai

With each new generation of Gemini models, the free tier will match or exceed the capability of the previous generation's paid Pro tier.

“as we go to the next generation you know it's like every time we've gone forward a generation then the sort of uh the new free tier is usually as good as the previous pro tier uh and sometimes better.” View on YouTube
Explanation

The pattern held for the next generation: Gemini's free tier now runs on Gemini 3 Flash, which benchmarks ahead of the prior generation's paid Gemini 2.5 Pro tier on most agentic tasks, while the new Gemini 3.1 Pro was moved behind a paywall.